In a major shift for online retail, Meesho announced that more than 75 percent of its platform orders now originate from personalized product feeds. This surge is powered by its proprietary AI recommendation engine, known as PRISM. The system represents a fundamental shift away from traditional, keyword-based search bars, catering instead to a growing demographic of users who prefer browsing over typing.
Serving an enormous user base of 264 million annual transacting users, PRISM analyzes real-time shopper behavior and intent signals. According to Meesho’s Chief Data Scientist, Debdoot Mukherjee, the next hundred million Indians coming online will discover products through browsing and speaking rather than typing. To support this, the PRISM engine runs on more than 100 AI ranking models and is trained on over 400 trillion input signals, executing an astonishing 6 trillion inferences every single day.
Furthermore, the system is designed specifically for “Bharat”—India’s diverse, non-metropolitan consumer base. It supports over ten languages, including Hindi, Bengali, Tamil, and Marathi. It also utilizes “Trendpulse,” a large language model that interprets regional fashion and shopping trends across local consumer clusters.
To keep operational costs low, Meesho runs these heavy AI workloads on its in-house machine learning infrastructure, the BharatMLStack. This localized tech stack achieves high-throughput processing at a fraction of conventional cloud costs. By accurately matching products with high-intent audiences, the AI engine is not only redefining the user experience for millions of shoppers but is also helping sellers scale their businesses with unprecedented efficiency.